Week 03 Assignment Covid

New types of data and new data science technologies enable new research. These new technologies are technologies such as the ability to combine existing data or the ability to generate synthetic data from existing knowledge. This week casus is based on such research. Data is generated by Synthea's COVID-19 module. The data was constructed using three peer-reviewed publications published in the early stages of the global pandemic, when less was known, along with emerging resources, data, publications, and clinical knowledge. The simulation outputs synthetic Electronic Health Records (EHR), including the daily consumption of Personal Protective Equipment (PPE) and other medical devices and supplies. The Data is stored in separate tables to avoid redundancy, with as a concequence that tables need to be combined and reorganized in dataframes for analysing purpose.

Keywords: merge data, subset data, clean data, generate data

You will learn about combining data with pandas and numpy and you will learn to visualize with bokeh. Concretely, you will preprocess the partly Synthetic Covid data in an appropiate format in order to conduct statistical and visual analysis. Learning objectives

• Combine multiple data sources for analysis
• Read, inspect, clean, reshape data
• Visualize data using bokeh
• Maintain development environment
• Apply coding standards and FAIR principles
• Reshape the dataset into a format suitable for visual and statistical analysis
• Use widgets to make the plot interactive
• Use GIS libraries to plot geographical data

Tutorials about combining data: https://github.com/fenna/BFVM22PROG1/blob/main/tutorials/tutorial_combine_data.ipynb

study case combining data:https://github.com/fenna/BFVM22PROG1/blob/main/study_cases/adults_who_binge_drank_in_hot_towns.ipynb

Please add the topics you want to learn about here: https://padlet.com/ffeenstra1/kzh2chaqleq3iovu

Your job is to visualize the lab values taken for COVID-19 patients of survived versus not survived patients.

The assignment consists of 6 parts:

• part 1: load the data
  • Exercise 1.1
• part 2: data wrangling
  • Exercise 2.1
• part 3: more wrangling
  • Exercise 3.1
• part 4: plot the data
  • Exercise 4.1
• part 5: plot patient location
  • Exercise 5.1

Part 1 and 4 are mandatory, part 5 is optional (bonus) Mind you that you cannot copy code without referencing the code. If you copy code you need to be able to explain your code verbally and you will not get the full score.

About the data

The data is generated by Synthea's COVID-19 module. The data was constructed using three peer-reviewed publications published in the early stages of the global pandemic, when less was known, along with emerging resources, data, publications, and clinical knowledge. The simulation outputs synthetic Electronic Health Records (EHR), including the daily consumption of Personal Protective Equipment (PPE) and other medical devices and supplies. For this assignment the conditions, patients, observations, careplans and encounters table will be used. The Data is stored in separate tables to avoid redundancy, with as a concequence that tables need to be combined and reorganized in dataframes for analysing purpose.

Source: Walonoski J, Klaus S, Granger E, Hall D, Gregorowicz A, Neyarapally G, Watson A, Eastman J. Synthea™ Novel coronavirus (COVID-19) model and synthetic data set. Intelligence-Based Medicine. 2020 Nov;1:100007. https://doi.org/10.1016/j.ibmed.2020.100007

Please download the data

Covid Patients

Patients are considered Covid patients if they are identified with CODE 840539006

Survivors

Patients that had covid and where tested negative after isolation have tested code 94531-1, SARS-CoV-2 RNA Pnl Resp NAA+probe (covid-sars test) + a value of Not detected (qualifier value). These patients are considered to be survived covid patients.

Non-Survivors

Patients that did not survived Covid have a DEATHDATE which is not null.

Lab values COVID-19 patients

Patients are monitored for blood and heart conditions once they are admitted in Hospital or under treatment. The lab values of interest are as follow:

• 48065-7 Fibrin D-dimer FEU [Mass/volume] in Platelet poor plasma
• 26881-3 Interleukin 6 [Mass/volume] in Serum or Plasma
• 2276-4 Ferritin [Mass/volume] in Serum or Plasma
• 89579-7 Troponin I.cardiac [Mass/volume] in Serum or Plasma by High sensitivity method
• 731-0 Lymphocytes [#/volume] in Blood by Automated count
• 14804-9 Lactate dehydrogenase [Enzymatic activity/volume] in Serum or Plasma by Lactate to pyruvate reaction

---

Part 1: Load the data (20 pt)

Instructions: Load the data of the following files. Preferably we read the data not with a hard coded data path but using a config file. See https://fennaf.gitbook.io/bfvm22prog1/data-processing/configuration-files/yaml

• conditions.csv
• patients.csv
• observations.csv
• careplans.csv
• encounters.csv

Get yourself familiar with the data. Create some meaningful overviews. Answer the following questions

1. How many patients are there
2. How many covid-patients are there
3. How many patients do have a 'Hospital admission for isolation' encounter

Hints

• use a unique dataframe for each file, use a meaningful name
• pandas.read_csv() method can be used to read a csv file
• pandas.DataFrame.head() method is often used to inspect the dataframe
• .unique() returns a list of unique values of a column

1.1 Code your solution

import pandas as pd
import numpy as np
from bokeh.io import output_notebook
import yaml
output_notebook()

Loading BokehJS ...

files = ['conditions.csv', 'patients.csv', 'observations.csv', 'careplans.csv', 'encounters.csv']
dfs = {}

with open("config.yaml", "r") as config:
    data_folder = yaml.safe_load(config)['data_folder']

    for file in files:
        dfs[file.split('.')[0]] = pd.read_csv(data_folder + file)

# Total number of patients
num_patients = len(dfs['patients'])

# Number of COVID patients
num_covid_patients = len(dfs['conditions'].loc[dfs['conditions']['CODE'] == 840539006])

# Number of admitted patients
num_admitted_patients = len(dfs['encounters'].loc[dfs['encounters']['DESCRIPTION'] == 'Hospital admission for isolation (procedure)'])

# Number of deceased patients
num_dead_patients = len(dfs['patients'][dfs['patients']['DEATHDATE'].notnull()])

1.2 Test your solution

The following function needs to be called. You can use this as a test. There are however more meaningful overviews you can create.

def part1(num_pat, num_cov, num_admitted, num_died):
    print(f'There are {num_pat} patients in total')
    print(f'There are {num_cov} covid patients')
    print(f'There are {num_admitted} admitted patients')
    print(f'{num_died} patients died')


part1(num_patients, num_covid_patients, num_admitted_patients, num_dead_patients)

There are 12352 patients in total
There are 8820 covid patients
There are 1867 admitted patients
2352 patients died

Expected outcome

There are 12352 patients in total There are 8820 covid patients There are 1867 admitted patients 2352 patients died

---

Part 2: Data Wrangling: set up the dataframe (30 pt)

In this part we are going to combine data to create a dataframe with values of interest for the lab values analysis.

We would like a dataframe containing the following information per record (only Covid patients!!!)

• PATIENT - the ID of the covid patient
• days - the number of days the patient is under observation
• CODE-Y - the code of the observation
• VALUE - the lab value of the observation

where only the following observation codes needs to be selected:

• 48065-7 Fibrin D-dimer FEU [Mass/volume] in Platelet poor plasma
• 26881-3 Interleukin 6 [Mass/volume] in Serum or Plasma
• 2276-4 Ferritin [Mass/volume] in Serum or Plasma
• 89579-7 Troponin I.cardiac [Mass/volume] in Serum or Plasma by High sensitivity method
• 731-0 Lymphocytes [#/volume] in Blood by Automated count
• 14804-9 Lactate dehydrogenase [Enzymatic activity/volume] in Serum or Plasma by Lactate to pyruvate reaction

The days information is not primarely available and needs to be calculated by substracting observation DATE - START.

An example of such a dataframe is given below:

days CODE-Y VALUE UNITS PATIENT 00079a57-24a8-430f-b4f8-a1cf34f90060 0.0 14804-9 241.8 U/L 00079a57-24a8-430f-b4f8-a1cf34f90060 0.0 2276-4 489.5 ug/L 00079a57-24a8-430f-b4f8-a1cf34f90060 0.0 48065-7 0.4 ug/mL 00079a57-24a8-430f-b4f8-a1cf34f90060 0.0 731-0 1.1 10*3/uL 00079a57-24a8-430f-b4f8-a1cf34f90060 0.0 89579-7 3.4 pg/mL 00079a57-24a8-430f-b4f8-a1cf34f90060 1.0 731-0 1.1 10*3/uL 00079a57-24a8-430f-b4f8-a1cf34f90060 2.0 14804-9 233.7 U/L 00079a57-24a8-430f-b4f8-a1cf34f90060 2.0 2276-4 497.3 ug/L 00079a57-24a8-430f-b4f8-a1cf34f90060 2.0 48065-7 0.3 ug/mL 00079a57-24a8-430f-b4f8-a1cf34f90060 2.0 731-0 1.0 10*3/uL

#Possible approach:

#Select all the patients with covid from the conditions table
#Combine conditions table (only covid patients) with the patient table into a covid_patient table
#select the only the relevant lab observations from the observations table into a lab_obs table
#merge the covid_patient table with the lab_obs table into a covid_patients_obs table
#clean the covid_patients_obs table (rename columns, select only relevant columns, sort, typecast, add days column)

Hints

• you can use pandas.DataFrame.merge() to merge dataframes
• df = df[(df.CODE == condition1 | df.CODE == condition1 )] selects rows with CODE of 2 conditional values
• df.DATE - df.START return days if DATE and START are datetime format
• pd.to_datetime() can be used to typecast to datetime

2.1 Code your solution

# Set correct date types for date columns
dfs['observations']['DATE'] = pd.to_datetime(dfs['observations']['DATE'], errors='coerce')
dfs['conditions']['START'] = pd.to_datetime(dfs['conditions']['START'], errors='coerce')
dfs['conditions']['STOP'] = pd.to_datetime(dfs['conditions']['STOP'], errors='coerce')

# Data frame of filtered COVID patients
filtered_covid_patients = dfs['conditions'].loc[dfs['conditions']['CODE'] == 840539006]

# Merge filtered COVID patients data frame with the observation data frame 
covid_patient = pd.merge(dfs['observations'], filtered_covid_patients, on='PATIENT')

# Data frame of filtered on the relevant observations
relevant_observations = ['48065-7', '26881-3', '2276-4', '89579-7', '731-0', '14804-9']
filtered_observations = covid_patient.loc[covid_patient['CODE_x'].isin(relevant_observations)]

# Create days column by subtracting the start and stop date
filtered_observations['days'] = (filtered_observations['DATE'] - filtered_observations['START']).dt.days

# Only select the columns that are desired
df = filtered_observations[['PATIENT', 'days', 'CODE_x', 'VALUE', 'UNITS']]

# Cast VALUE column to float
df['VALUE'] = pd.to_numeric(df['VALUE'], errors='coerce')

df.head()

/tmp/ipykernel_4001173/1333209855.py:17: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  filtered_observations['days'] = (filtered_observations['DATE'] - filtered_observations['START']).dt.days
/tmp/ipykernel_4001173/1333209855.py:23: SettingWithCopyWarning: 
A value is trying to be set on a copy of a slice from a DataFrame.
Try using .loc[row_indexer,col_indexer] = value instead

See the caveats in the documentation: https://pandas.pydata.org/pandas-docs/stable/user_guide/indexing.html#returning-a-view-versus-a-copy
  df['VALUE'] = pd.to_numeric(df['VALUE'], errors='coerce')

	PATIENT	days	CODE_x	VALUE	UNITS
273	f58bf921-cba1-475a-b4f8-dc6fa3b8f89c	0	731-0	1.1	10*3/uL
292	f58bf921-cba1-475a-b4f8-dc6fa3b8f89c	0	48065-7	0.4	ug/mL
293	f58bf921-cba1-475a-b4f8-dc6fa3b8f89c	0	2276-4	332.4	ug/L
294	f58bf921-cba1-475a-b4f8-dc6fa3b8f89c	0	89579-7	2.3	pg/mL
295	f58bf921-cba1-475a-b4f8-dc6fa3b8f89c	0	14804-9	223.9	U/L

---

Part 3: Data Wrangling, split into survived and not survived (10 pt)

Now we have the required data we would like to split the data into survived and not survived. First we fetch all the ids of the survived and deceased patients. We can use these ids to select the records of the survived patients and the patients that did not survived

Your job is to split the data into survived and not survived records. There are multiple ways to do this. One way is the .isin() method

# Store the data in variables with the expected names, since I stored all the datasets in a dictionairy
care_plans = dfs['careplans']
observations = dfs['observations']
patients = dfs['patients']

#the following code is given, RUN THIS CELL
#get survived and deceased ids
completed_isolation_patients = care_plans[(care_plans.CODE == 736376001) & (care_plans.STOP.notna()) \
                                          & (care_plans.REASONCODE == 840539006)].PATIENT
negative_covid_patient_ids = observations[(observations.CODE == '94531-1') \
                                          & (observations.VALUE == 'Not detected (qualifier value)')].PATIENT.unique()
survivor_ids = np.union1d(completed_isolation_patients, negative_covid_patient_ids)
deceased_ids = patients[patients.DEATHDATE.notna()].Id

3.1 Code your solution

# Filter the survived and deceased patients
survived = df.loc[df['PATIENT'].isin(survivor_ids)]
deceased = df.loc[df['PATIENT'].isin(deceased_ids)]

num_survived = len(survived)
num_deceased = len(deceased)

3.2 Test your solution

def test3(survived, died):
    print(f'patients records survived: {survived}, patients records deceased {died}')

test3(num_survived, num_deceased)

patients records survived: 57303, patients records deceased 16793

Expected outcome

patients records survived: 57303, patients records deceased 16793

---

Part 4: Plot the data (20 pt)

Create plots with the lab data, for each code one plot. Separate the survivors and the deceased by color. An example of such a plot is given below. You can create 6 plots in one grid (for each code one plot) or use a widget (for instance a drop down menu widget) to select a lab CODE. Plot on the x-axis the days, on the y-axis the VALUE. Use proper labels, titles and legends.

4.1 Code your solution

from bokeh.plotting import figure, show, ColumnDataSource
from bokeh.layouts import gridplot
from bokeh.models import Legend

def make_plot(code, title, y_label):
    '''Function that returns a scatter plot for the passed data'''

    sur = survived.loc[survived['CODE_x'] == code]
    dec = deceased.loc[deceased['CODE_x'] == code]
    p = figure(width=490, height=400, title=title, x_axis_label='Days', y_axis_label=y_label)
    sur_scatter = p.circle(x=sur['days'], y=sur['VALUE'], size = 5, color="green")
    dec_scatter = p.circle(x=dec['days'], y=dec['VALUE'], size = 5, color="red")
    
    legend = Legend(items=[
        ('Survived',   [sur_scatter]),
        ('Deceased', [dec_scatter]),
    ])

    p.add_layout(legend, 'right')

    return p
    
    
def make_plot_array(plots):
    '''Function that creates an array of scatter plots for the desired codes'''

    codes = {
        '48065-7': ['Fibrin D-dimer FEU [Mass/volume] in Platelet poor plasma', 'ug/mL'], 
        '26881-3': ['Interleukin 6 [Mass/volume] in Serum or Plasma', 'pg/mL'],
        '2276-4': ['Ferritin [Mass/volume] in Serum or Plasma', 'ug/mL'],
        '89579-7': ['Troponin I.cardiac [Mass/volume] in Serum or Plasma by High sensitivity method', 'pg/mL'], 
        '731-0': ['Lymphocytes', '10*3/uL'], 
        '14804-9': ['Lactate dehydrogenase [Enzymatic activity/volume] in Serum or Plasma by Lactate to pyruvate reaction', 'U/L'],
    }
    
    for k, v in codes.items():
        plots.append(make_plot(k, v[0], v[1]))
 
    
plots = []
make_plot_array(plots)
show(gridplot(plots, ncols=2))

Part 5: Plot the location of the patients (10 pt)

This is a bonus part. Can you plot the patients location on a map? See also https://docs.bokeh.org/en/latest/docs/user_guide/geo.html

You can use either package folium or geopandas. You need the Latitude and Longitude information from the patient tabel

5.1 Code your solution

# For this assigment I followed a tutorial and copied code parts from it
# The tutorial can be found on: https://towardsdatascience.com/creating-an-interactive-map-in-python-using-bokeh-and-pandas-f84414536a06

def x_coord(x, y):
    '''Function that converts lat/lon coordinates to mercator coordinates'''
    
    lat = x
    lon = y
    
    r_major = 6378137.000
    x = r_major * np.radians(lon)
    scale = x/lon
    y = 180.0/np.pi * np.log(np.tan(np.pi/4.0 + 
        lat * (np.pi/180.0)/2.0)) * scale
    return (x, y)


dfs['patients']['coordinates'] = list(zip(dfs['patients']['LAT'], dfs['patients']['LON']))
mercators = [x_coord(x, y) for x, y in dfs['patients']['coordinates']]
dfs['patients']['mercator'] = mercators
dfs['patients'][['mercator_x', 'mercator_y']] = dfs['patients']['mercator'].apply(pd.Series)

p = figure(
    x_axis_type="mercator", y_axis_type="mercator", width=800, title='Patient locations')
p.add_tile("CartoDB Positron", retina=True)

p.circle(x="mercator_x", y="mercator_y",
         size=2,
         fill_color="dodgerblue", line_color="dodgerblue",
         fill_alpha=0.3,
         source=dfs['patients'])

show(p)